Toward Free‐Breathing Renal CEST MRI: Preclinical Evaluation of a Novel Prospective Respiratory Triggering Approach

ABSTRACT

Purpose

Endogenous renal chemical exchange saturation transfer (CEST) imaging enables contrast‐agent‐free assessment of renal metabolism but is highly sensitive to respiratory motion. Existing approaches, such as timed breathing, require high patient compliance or preclinical mechanical ventilation, limiting clinical translation. We evaluated a free‐breathing prospective respiratory triggering method that allows continuous saturation during active breathing while adaptively aligning image acquisition with the expiratory phase.

Methods

Prospective triggering was based on a modified real‐time respiratory signal, with a temporal shift continuously updated according to the current respiratory rate. Simulations assessed its performance across saturation durations of 1–6 s, respiratory rates representative of preclinical and clinical settings (f = 8–90 bpm), and varying respiratory frequency (CV = 0%–25%). Results were compared to conventional real‐time triggering. For in vivo validation, prospective triggering was implemented on a preclinical 7 T MRI system using a Raspberry Pi‐based hardware and software. Six mice underwent CEST imaging with saturation durations of 1, 3, and 5 s using both real‐time and prospective triggering, and the Mean Squared Error of Lorentzian fit was used to compare the two approaches.

Results

The real‐time trigger error shows a periodic dependence on respiratory frequency, which diminishes with increasing variability, whereas the prospective trigger reduces this periodicity. In vivo, prospective triggering significantly (p < 0.05) reduced motion‐induced scattering in the Z‐spectrum for saturation times up to 3 s and enabled the detection of distinct metabolic contrasts across renal compartments.

Conclusion

Prospective triggering effectively reduces motion artifacts in preclinical renal CEST imaging.

1. Introduction

Endogenous renal chemical exchange saturation transfer (CEST) imaging is a promising MRI‐based technique for assessment of renal metabolic processes [1]. CEST imaging relies on the acquisition of a series of images with varying saturation offsets. Based on the signal intensity of a certain pixel in this image series, a so‐called Z‐spectrum is generated, from which metabolic information can be derived. The examination requires the kidney to remain spatially stable throughout the image acquisition. Otherwise, signal intensities from different tissue locations are combined, leading to data scattering [2, 3] and metabolic misinterpretation [4].

Respiratory motion thus represents a challenge in abdominal CEST imaging and has been addressed by timed‐breathing protocols in various studies, in which subjects synchronize their breathing with the image acquisition [3, 5, 6, 7]. This approach has allowed us to distinguish between healthy and pathological renal tissue [6] and the investigation of metabolic differences in native kidneys [3]. These studies demonstrate the feasibility of abdominal CEST imaging and also suggest that saturation can be applied throughout the active respiratory cycle, while the image needs to be acquired during the expiratory phase. This observation provides a conceptual basis for the development of novel respiratory motion correction strategies, which are sought in abdominal CEST studies [3, 5, 8, 9, 10].

Preclinical studies further highlight the potential of renal CEST imaging for assessing acute kidney injury (AKI) [11, 12, 13], ischemia [14], chronic kidney disease [15], diabetic kidney disease [10], acute renal allograft rejection [16] and pH‐mapping [17, 18, 19, 20]. However, translating these findings into clinical practice remains challenging, as preclinical studies typically rely on mechanically ventilated animals [11, 17], an approach that is difficult to justify in clinics. Clinical timed‐breathing protocols require high patient compliance and may be unsuitable for individuals prone to dyspnea.

In both preclinical and clinical MRI, respiratory triggering is commonly used but is suboptimal for CEST imaging because the required saturation period often spans over multiple breathing cycles, which hinders reliable synchronization with expiration (Figure 1a). Although continuous saturation [8, 21], shortened saturation duration [9] or fixed temporal trigger shifts [10] can partially address this issue, these approaches either limit the saturation strategy or may fail for longer saturation durations and respiratory variability.

FIGURE 1.

Based on the real‐time trigger (a, top), the temporal shift Δt is calculated from the first three respiratory periods (T₁, T₂, T₃) of the real‐time trigger for prospective trigger (a, bottom). The temporal deviation between the center of the image acquisition (label) and the nearest center of the exhaled state (target) is then calculated for both the real‐time trigger and the prospective trigger. The error is scaled relative to the two real‐time trigger events from 0 to 1, where 0 indicates perfect acquisition in the exhaled state and 1 indicates maximal error (acquisition in the inhaled state). The example in (a) is a snapshot of the simulation with an average respiratory period T of 60 bpm, coefficient of variation CV of 12.5%, and a saturation time of 1 s, resulting in $\epsilon$ _RT of 0.96 and $\epsilon$ _PT of 0.03. For in vivo evaluation, a TTL‐based trigger signal was generated from the mouse respiratory profile to represent the real‐time trigger (b). For prospective triggering, this real‐time TTL signal was processed by a Raspberry Pi and transmitted to the scanner as the prospective trigger.

Therefore, we proposed a novel triggering approach based on a modified real‐time respiratory signal, in which the temporal shift is continuously updated to the current respiratory rate, allowing saturation throughout the active breathing cycle while prospectively aligning image acquisition with expiration (Figure 1). This prospective triggering was evaluated in simulations across different saturation durations, clinical and preclinical respiratory rates and variabilities, and compared to conventional real‐time triggering. Prospective trigger was implemented on a preclinical 7 T MRI system and validated in vivo on an animal model.

2. Methods

2.1. Prospective Trigger

The prospective trigger signal represents a temporally modified version of the real‐time respiratory trigger. Trigger events are shifted so that the center of the image acquisition coincides with the middle of the expiration phase, while accounting for both the saturation duration and acquisition time. The temporal shift applied to each trigger event is continuously updated based on the current respiratory period, estimated from the average of the three most recent real‐time trigger intervals, thereby increasing robustness to respiratory variability. The temporal offset Δt applied to each real‐time trigger event is calculated by

$$\mathrm{\Delta }t=\left(n·\bar{T}-t_{\mathit{sat}}\right)+\frac{\bar{T}-t_{\mathit{acq}}}{2},\text{with}n=⌈\frac{t_{\mathit{sat}}}{\bar{T}}⌉$$ (1)

where $\bar{T}$ is the respiratory period calculated as the mean of the three most recent trigger intervals, t _sat is the saturation duration, and t _acq is the acquisition time. The first addend shifts the acquisition prospectively to the beginning of the n‐th respiratory cycle, calculated by the ceiling operator ⌈·⌉, while the second addend centers the acquisition within the expiration phase (Figure 1a).

2.2. In Silico Simulation

To evaluate CEST acquisition timing and compare real‐time and prospective triggering under identical respiratory conditions, a simulation framework was implemented. Each simulated acquisition was based on a respiratory scenario defined by real‐time trigger events $\mathit{RT}=\left(r_i\right)$, where each $r_i$ corresponds to maximal inhalation (Figure 1). The sequence was generated recursively as $r_i=r_{i-1}+T_i$, with breathing periods $T_i$ drawn from a lognormal distribution [22] using numpy.random.lognormal(${\mu }_{\mathit{ln}},{\sigma }_{\mathit{ln}}$) to model respiratory variability, with the expected mean value ${\mu }_{\mathit{ln}}$ and standard deviation ${\sigma }_{\mathit{ln}}$. The distribution parameters ${\mu }_{\mathit{ln}}$ and ${\sigma }_{\mathit{ln}}$ were derived from the mean respiratory period $T$ and the coefficient of variation (CV) by ${\sigma }_{\mathit{ln}}=\sqrt{\ln \left(1+{\mathit{CV}}^2\right)}\text{and}{\mu }_{\mathit{ln}}=\ln (T)-\frac{1}{2}{\sigma }_{\mathit{ln}}^2$.

To calculate the prospective triggering acquisition timing, the temporal shift Δt was calculated using Equation (1) based on the saturation duration (t_sat), acquisition time (t_acq), and the respiratory period $\bar{T}$ estimated by averaging the first three breathing periods. In real‐time triggering, acquisition timing was calculated directly to the real‐time trigger event. Timing accuracy was quantified by the error $\epsilon$, defined as the normalized difference between the acquisition center and the nearest exhaled state (Figure 1a). Simulations covered breathing rates from 8 bpm (resting humans) to 90 bpm (lightly anesthetized B6 mice [23]) with CV ranging [24] from 0 to 0.25.

Acquisition time was fixed at 131 ms, and the number of simulated frequency offsets was set to 73, consistent with the optimized in vivo CEST sequence, while saturation duration was varied between 1 and 6 s.

2.3. Hardware and Software Adaptions for Implementing Prospective Triggering

The TTL trigger signal from the breathing monitoring system was processed via the GPIO interface of a Raspberry Pi, where the temporally shifted trigger was computed and forwarded to the MRI scanner. To minimize latency, a multithreaded software architecture was used. One thread continuously acquired trigger events, estimated the respiratory period by averaging the last three trigger intervals, and computed the temporal offset Δt (Equation 1), while a second thread generated the delayed TTL output.

2.4. In Vivo Experiments

In vivo experiments were performed on a 7 T preclinical MRI system (MRS*Drymag 7017PW, MR Solutions, Guildford, UK) with a mouse body coil. For each scan, RF calibration with a stimulated‐echo method [25, 26] and manual B₀ shimming were performed.

Morphological images were acquired using a fast spin‐echo sequence for CEST planning and segmentation (coronary FOV 20.0 × 10.0 × 1.5 mm [3], matrix size of 128 × 128, TE/TR = 6.0/5000 ms, turbofactor TF = 32, 3 averages, acquisition bandwidth 33.3 kHz). For CEST imaging, the matrix was reduced to 64 × 64 (TE/TR = 4.09/5500 ms, TF = 32, one average). Each CEST image acquisition was divided into two CEST‐sequence blocks, each independently triggered to distinct exhalation phases and acquiring half of k‐space within 131 ms. The TR was defined according to literature recommendations [19]. In the context of respiratory triggering, TR denotes the minimum interval between the start of consecutive CEST‐sequence blocks. Saturation was applied using a 1.5 μT block pulse with durations of 1, 3, and 5 s. Z‐spectra were acquired at 73 frequency offsets (±1800 Hz/˜6.0 ppm, 50 Hz/˜0.17 ppm steps), resulting in a protocol duration of 13 min and 23 s (˜14.5 min with real‐time or prospective triggering). A real‐time triggered image without saturation was acquired for normalization.

Six healthy adult female B6 albino mice were scanned using both real‐time and prospective triggering for t _sat = [1, 3, 5] s. For reproducibility assessment, six healthy female BALB/c mice were scanned twice within the same imaging session using prospective triggering, with identical shimming and positioning, with t_sat of 3 s. Anesthesia was induced with 3.0% and maintained with 1.75%–2.0% isoflurane (Isothesia, Covetrus, Portland, USA) in air, maintaining respiratory rates within 40–60 bpm. Body temperature was kept at 37°C. Respiration was monitored using a small‐animal gating system (ERT Control/Gating Module, Model 1030, SA Instruments Inc., New York, USA), generating a TTL trigger signal that was either directly applied for real‐time triggering or routed through a Raspberry Pi for prospective triggering (Figure 1b).

All animal experiments were approved by the Laboratory Animal Care and Use Committee of the First Faculty of Medicine, Charles University, and the Ministry of Education, Youth and Sports of the Czech Republic (MSMT‐16533/2024‐5). The approved protocol was in accordance with the Act of the Czech Parliament for the Protection of Animals Against Cruelty No. 246/1992 and the Directive 2010/63/EU of the European Parliament.

2.5. Postprocessing

Data analysis and visualization were performed using custom Python scripts. For CEST analysis, the renal inner medulla (IM), outer medulla (OM), and cortex (CO) were manually segmented using ITK‐SNAP [27] and the kidney was cropped across the image series for an Image Downsampling Expedited Adaptive Least‐squares (IDEAL) fitting framework [18, 28] with a 5‐pool Lorentzian model [29],

$$\widehat{Z}(\omega )=Z_{\mathit{\text{base}}}-\sum _i^{\mathit{\text{Pool}}}{L}_i(\omega ),\text{with}{L}_i(\omega )=A_i·\frac{\frac{{\mathrm{\Gamma }}_i^2}{4}}{\frac{{\mathrm{\Gamma }}_i^2}{4}+{(\omega -\mathrm{\Delta }\omega )}^2}$$ (2)

where $A$ is the amplitude, $\mathrm{\Gamma }$ is the full‐width at half‐maximum (FWHM), $\mathrm{\Delta }\omega$ is the frequency offset of the CEST‐pool.

The cropped image series were spatially downsampled by averaging kernel using patch sizes of [1 × 1], [2 × 2], [4 × 4], [8 × 8], [16 × 16], and [32 × 32] pixels. For each patch, B₀ inhomogeneity was estimated from the minimum of a spline‐interpolated Z‐spectrum [30] (scipy.interpolate.UnivariateSpline() [31], smoothing factor s = 0.01, interpolation step size = 0.01 ppm) and used as a fixed parameter for fitting.

The fitting was performed by nonlinear least‐squares optimization using the Trust‐Region‐Reflective algorithm (scipy.optimize.curve_fit() [31]) with initial values and bounds for water (A = [0.02;0.9;1.0], $\mathrm{\Gamma }$ = [0.3;1.4;10.0]), amine (A = [0.0;0.01;0.2], $\mathrm{\Gamma }$ = [1.0;1.5;3.5], $\mathrm{\Delta }\omega$= [1.0;2.2;2.5]), amide (A = [0.0;0.025;0.2], $\mathrm{\Gamma }$ = [0.4;0.5;3.0], $\mathrm{\Delta }\omega$ = [3.0;3.5;4.0]), nuclear Overhauser effect (NOE) (A = [0.0;0.02;0.4], $\mathrm{\Gamma }$ = [1.0;3.0;5.0], $\mathrm{\Delta }\omega$ = [−4.5;‐3.5;2.0]), magnetization transfer (MT) (A = [0.0;0.1;1.0], $\mathrm{\Gamma }$ = [10.0;25.0;100.0], $\mathrm{\Delta }\omega$ = [−4.0;‐2.0;‐2.0]), and baseline ($Z_{\mathit{\text{base}}}$ = [0.5; 1.0; 1.0]) at 7 T were adopted from Windschuh et al. [29] In the IDEAL method, amplitude and FWHM were constrained to ±10% and frequency offsets to ±5% of the initial fit results [28].

As no fat suppression was applied and lipid signals may influence the NOE effect [32] around −3.5 ppm, the signal observed at this offset is referred to as pseudo‐NOE (pNOE) [33] in this study.

2.6. Statistics

Previous studies [2, 3] demonstrated that motion artifacts increased data‐scattering in the Z‐spectrum. Therefore, the mean squared error (MSE) between measured $Z\left({\omega }_i\right)$ and Lorentzian‐fitted $\widehat{Z}(\omega )$ spectra with 73 frequency offsets for each pixel were assessed to quantify motion in CEST data:

$$\mathit{MSE}=\frac{1}{73}\sum _{i=1}^{73}{\left(Z\left({\omega }_i\right)-\widehat{Z}\left({\omega }_i\right)\right)}^2$$ (3)

ROI‐averaged MSE values for real‐time and prospective trigger were compared using the Wilcoxon signed‐rank test for each saturation duration.

For each CEST pool, the fitted amplitudes were averaged within the respective ROIs. The normality of the pairwise differences in ROI‐averaged amplitudes was assessed using the Shapiro–Wilk test (α = 0.01). Paired t‐tests with Bonferroni correction were performed to determine significant metabolic differences between the IM, OM, and CO for prospective triggering. In reproducibility experiments, fitted amplitudes were averaged within the respective ROIs and compared between measurements using intraclass correlation coefficient ICC(3,1) [34] and categorized as poor (< 0.5), moderate (< 0.75), good (< 0.9), or excellent (> 0.9). The MSE of the initial feasibility measurement using real‐time triggering was compared with the MSE from the reproducibility measurements using a Mann–Whitney test. Significance levels were marked by (****) for p ≤ 0.0001, (***) for p ≤ 0.001, (**) for p ≤ 0.01, (*) for p ≤ 0.05; p > 0.05 was considered as nonsignificant. All statistical analyses were performed using the Python library SciPy [31, 35].

3. Results

3.1. In Silico Simulation

The error of the real‐time trigger exhibited a pronounced periodic dependence on respiratory frequency and reached its maximum at respiratory periods corresponding to multiples of the saturation time plus half of the acquisition time (Figure 2). With increasing variability of the respiratory frequency, the periodicity of the error decreased; this effect occurred earlier at higher respiratory frequencies than at lower frequencies. For the prospective trigger, the periodic dependence of the error was reduced, showing only minor stepwise increases in error at positions corresponding to the maxima observed for the real‐time trigger.

FIGURE 2.

Simulated averaged error $\bar{\mathrm{\epsilon }}$ of the real‐time trigger and the prospective trigger in positioning image acquisition into the exhaled state. An error value of 1 indicates that image acquisition always occurs during the inhaled state, whereas an error value of 0 indicates that image acquisition always occurs during the exhaled state.

3.2. In Vivo Experiments

Reduced motion was observed in the CEST image series with prospective triggering (Supporting Information, Video S1). Prospective triggering significantly (p < 0.05) reduced the MSE in the kidney (Figure 3a) compared to real‐time triggering for saturation durations of 1 s (from (4.96 ± 5.48) × 10⁻³ to (0.62 ± 0.21) × 10⁻³) and 3 s (from (1.09 ± 0.43) × 10⁻³ to (0.51 ± 0.28) × 10⁻³). No significant difference (p > 0.05) in MSE was found between prospective ((0.93 ± 0.88) × 10⁻³) and real‐time triggering ((0.61 ± 0.51) × 10⁻³) for 5 s saturation. An exemplary Z‐spectrum from a cortical pixel (Figure 3b) showed pronounced scattering between 6.0 and 1.0 ppm (Figure 3c), resulting in an increased MSE of 1.07 × 10⁻³. In contrast, the same pixel acquired with prospective triggering exhibited reduced scattering (Figure 3d) and a lower MSE of 0.05 × 10⁻³. Additionally, the Z‐spectrum obtained with prospective triggering displayed lower signal intensities at the spectral flanks and broader fitted pools compared to real‐time triggering.

FIGURE 3.

The mean MSE of the Lorentzian fit (a) for individual mice (each shown in a different color) was significantly (p < 0.05) reduced in the in vivo experiments for saturation times of 1 and 3 s, when using prospective triggering compared to real‐time triggering. In the segmentation process (b), a margin from the kidney boundary was maintained, particularly at borders perpendicular to the cranial‐caudal axis (frequency encoding direction), to minimize signal changes induced by water‐fat shift artifacts that could affect the IDEAL fitting results. A representative Z‐spectrum with a saturation time of 3 s from a single voxel in the segmented cortex exhibits pronounced data scatter and increased fitting residuals with real‐time triggering (c). In contrast, the Z‐spectrum from the same voxel acquired with prospective triggering shows a smoother spectral profile and reduced residuals (d).

B₀ maps were generally homogeneous for both triggering approaches but appeared more spatially continuous with prospective triggering (Figure 4). The Lorentzian fit MSE decreased from (0.85 ± 2.71) × 10⁻³ with real‐time triggering to (0.12 ± 0.09) × 10⁻³ with prospective triggering. CEST contrast maps acquired prospectively showed clearer renal structure delineation, with medulla‐cortex differentiation visible in the amine, pNOE, and MT maps, whereas such features were less discernible in real‐time triggered data. Structural contrast was less pronounced in the amide maps. Localized cluster‐like variations were observed in the lower pole of the left kidney in amine maps and the upper pole of the right kidney in MT maps for prospective‐triggered images (Figure 4).

FIGURE 4.

Exemplary B₀ maps (left), Lorentzian fit MSE maps (second from left), and corresponding CEST contrast maps from in vivo measurements acquired with real‐time triggering (top row) and prospective triggering (bottom row) at a saturation duration of 3 s. For the exemplary maps, mean CEST amplitudes (mean ± SD) in inner medulla, outer medulla, and cortex were in corresponding order: amine: (8.79 ± 1.21) %, (7.81 ± 1.40) %, (8.91 ± 1.90) %; amide: (10.12 ± 1.33) %, (9.71 ± 1.11) %, (9.13 ± 1.20) %; pNOE: (15.36 ± 1.50) %, (16.28 ± 1.77) %, (18.94 ± 2.23) %; MT: (2.82 ± 0.24) × 10⁻³%, (3.13 ± 0.35) × 10⁻³%, (3.46 ± 0.36) × 10⁻³%.

The Shapiro–Wilk test confirmed normal distribution of the data acquired with prospective triggering. Significant differences in CEST effects were observed between renal compartments (Figure 5). For all investigated saturation durations, the pNOE CEST effect differed significantly between IM and CO, with an additional significant difference between IM and OM at 1 s. At a saturation duration of 3 s, significant differences in the amine CEST effect were observed between IM and OM, whereas at 5 s, the MT pool showed significant differences between IM and both OM and CO.

FIGURE 5.

ROI‐averaged CEST effects from the in vivo experiments. Statistical significance was assessed using a paired t‐test with Bonferroni correction, with significance levels defined as (**) for p ≤ 0.01 and (*) for 0.01 < p ≤ 0.05. Box plot shows the median; dots represent outliers.

MSE was significantly reduced (p < 0.05) compared to real‐time triggering, with MSE of (0.33 ± 0.17) × 10⁻³ and (0.49 ± 0.30) × 10⁻³ for the first and second reproducibility measurements, respectively. Reproducibility was poor for amide and NOE CEST effects (ICC(3,1) < 0.5) and for amine in the IM (ICC(3,1) = 0.47), moderate for MT effects (0.55 < ICC(3,1) < 0.66), and good for amine in the OM (ICC(3,1) = 0.76) and cortex (ICC(3,1) = 0.85).

4. Discussion

The prospective trigger was successfully evaluated in simulations, preclinical renal CEST imaging, and compared to conventional real‐time triggering.

In silico, the error of the prospective trigger increased stepwise at respiratory periods in which real‐time triggering performed worse. This behavior arises from the ceiling function in Equation (1), which shifts acquisition to the next respiratory cycle and thereby increases sensitivity to respiratory variability, as indicated by simulations at longer saturation times. Simulations show that prospective triggering reduces the influence of respiratory rate on successful image acquisition.

The quality of Z‐spectra was significantly improved using the prospective trigger in vivo for saturation times up to 3 s. For a saturation time of 5 s, no significant improvement was observed. This lack of significance is likely due to the variation in respiratory frequency over longer saturation times, which may cause the calculation of Δt based on the previous three breaths to fail in accurately predicting the duration of subsequent multiple breaths. In addition, the relative error between the two triggering approaches depends on the degree of synchrony between respiration and image acquisition, which may change during the measurement, as suggested by the Z‐spectra in Figure 3c, where motion‐induced data scatter is observed within a specific interval only. This effect has been investigated in vitro [3] and may contribute to the inter‐individual variability observed in the MSE. Saturation times over 4 s are mainly used in exogenous CEST imaging [12, 14, 16, 17], thus the applicability of prospective triggering may be limited in these protocols.

Increased CEST effects at 2.2 ppm in the IM and OM, potentially related to amine groups [36], are consistent with findings by Wang et al. [10] In contrast, amide signals showed no clear compartment‐specific pattern, in agreement with Zhang et al. [13] and Tao et al. [11], who investigated APT effects in AKI. Zhang et al. also showed a corticomedullary increase in rNOE toward the CO. However, the use of fat suppression in their study, in contrast to the present work, may explain the differences in the observed pNOE CEST effect [36].

Our study has limitations that are important to note. First, the CEST results depend on saturation parameters [37] and reflect a metabolic snapshot that may vary with hydration [38], anesthesia [39], circadian factors [40], and respiration‐dependent recovery times may influence the z‐spectra. In addition, arterial pulsation, peristaltic motion (Supporting Information, Video S1), and B₁/B₀ field alterations during respiration may contribute to the cluster‐like patterns observed in the CEST maps, likely arising from locally varying saturation efficiency and overly constrained fit boundaries in the IDEAL fitting process (e.g., amine map at the renal pole in Figure 2). Second, the reproducibility was limited, particularly for pNOE and amide, whereas MT showed predominantly moderate and amine largely good reproducibility. The observed differences may be attributed to temporal metabolic variations in the kidney, potentially related to anesthesia [39, 41, 42, 43, 44, 45], which hampers access to reproducibility measurements under consistent metabolic conditions. However, metabolic variability is unlikely to compromise the feasibility of prospective triggering, as MSE was reduced compared to real‐time acquisitions, including in reproducibility experiments. Future studies may further improve robustness by incorporating B₁/B₀ inhomogeneity correction methods [5, 46], saturation pulses less sensitive to field alterations [47], and assessing reproducibility in vitro using breathing phantoms under controlled metabolic conditions [3]. In vivo, residual abdominal motion may be mitigated by fasting or butylscopolamine [48] administration and further addressed using image‐based motion correction methods [49]. Third, the simulations represent a simplified model of respiratory triggering in CEST imaging, whereas clinical respiration may be more complex and irregular [24, 50], which has so far only been investigated in simulations.

Thus, future studies could validate prospective triggering in clinical in vivo settings. Implementation may be achieved through hardware‐based solutions, as demonstrated here, or software integration. Vendor‐independent frameworks such as Pulseq [37], which support external triggers, may further enable simulation‐driven optimization of triggering and saturation parameters for renal CEST imaging [37, 47, 51].

5. Conclusion

Prospective triggering effectively reduces motion artifacts in preclinical renal CEST imaging under free‐breathing conditions.

Funding

This work was supported by Jürgen Manchot Stiftung, the Univerzita Karlova v Praze (SVV 260 635/2025), the Deutsche Forschungsgemeinschaft (530863408, NE 2136/3‐1), the Ministerstvo Školství, Mládeže a Tělovýchovy (LM2023050), and the European Regional Development Fund (CZ.02.1.01/0.0/0.0/18_046/0016045). Julia Stabinska was supported by the National Institutes of Health (K99 DK138294).

Conflicts of Interest

Ruslan Garipov is an employee of MR Solutions.

Supporting information

Video S1: This video compares a CEST image series acquired using prospective triggering and real‐time triggering.

Data S1: Program code.

Data Availability Statement

The Python code ProspectiveTrigger.py for computation of the trigger shift and generation of a prospective trigger signal is available in the Supporting Information. Other data are available upon request.